Before You Tell Me "32-Bit Float" Removes Clipping Or Makes the Pre-Amp "Gain Knob" Obsolete
HTML-код
- Опубликовано: 9 фев 2025
- This is a video I'll point viewers to who want to argue that 32-Bit Float must work because they bought a device because they think it does.
I explain why 32-bit-float has nothing to do with mic gain and clipping
maxrottersman....
Hilarious. You deny it’s benefits for all but the last 10 or 20 seconds. Oh yeah, and you get 8 additional bits of dynamic range, which is exactly why it’s identified as 32 bit float which has been used in software several decades now.
It has benefits in post production. It has no benefits in analog capture. You get 8 additional bits of range, but at a cost of "dynamic" fidelity. Not that it matters, a Focusrite Scarlett, for example, can't resolve past 0.3uV. When a Shure SM58's output dips below 0.000000452 volts for example, it's just noise. That means that you can only effectively resolve to 14 bits (and I'd say that's a stretch). Yes, 32-bit float has been in use for decades. But NOT for analog to digital quantization. I'll go into this in greater detail, and with tests, in the coming weeks. These will be tests you can replicate, like any scientific test.
You want to believe 32-bit float works because you don't want to believed you've been cheated. But you have. Sorry. Trying to find something in my argument that is "wrong" won't change the conclusion. Cheers!
@ Good luck. But, what’s the point if it does provide the intended benefit?
@@JamesGrieve-kt4su Every benefit has a tradeoff. There are few free lunches in this world ;) The point is, what if you're making a trade-off you don't want? Most manufacturers market the benefits, not the trade-offs. This is true for everything. So caveat emptor--the buyer has to figure out what the trade-offs are. In this case, the manufacturers imply that the trade-off is more memory storage. You need to save 25% more data to reduce clipping. Logically, that makes sense. Most people don't think past that. We WANT to believe new technology is going to make us better. So a lot of stuff going on here.
Anyway, the point could be "that" intended benefit would give you poorer audio fidelity as the high end. That is, forget the points where it clips, the amplitudes are being recorded unproportionally (flattened and extended at same time). This might not be noticeable to most people--perhaps no one in random audio. But if you studied the waveform you would notice it. The bigger problem is noise at the other end. You're pulling it up. You're right, it all depends on if the benefit is good for you. But you want the intended benefit to be less clipped audio at the expense of more memory used. I'm saying that you don't get that benefit. You get the benefit of less clipped audio through signal processing or pre-amp limiter circuitry. Both adulterate the signal. They have nothing to do with a "floating" representation (mantissa^exponent) of numerical values. It sounds like you have an open mind. When the light bulb goes off for you you'll have a good laugh! I find it deeply fascinating!!! I suspected it was a fraud in the beginning, but it took me around 100 hours to really figure out what does what. Again, fascinating stuff. Difficult to explain!
@ If you proceed with your test please use transducers with high signal to noise ratio and high sensitivity, and or both qualities in separate microphones, pick-ups, etc.
In my experience the noise floor of the recorders is negligible - such that significant gain is necessary to make it an issue. Recording is optimized by proper sensor placement such that the signal is noticeably greater than the noise. Thus while the noise floor is always greater than zero, 32 bit float recording in practice provides higher performance at the upper limits of the sensors range when exceptional sensors are used.
The additional memory used was a 20th century challenge. It’s minuscule in present day data storage scenarios, especially in AV production professions.
@@JamesGrieve-kt4su You're putting me to work here! My focus is microphones. But I'm happy to include other kinds of transducers. A transducer produces a fluctuating voltage that must be amplified. Let's start there. Tell me what the voltage is of your transducer just above noise and what it is before max SPL. Tell me how much audio db range between those voltage. Then we can start thinking through the whole process. Also, can you run Python scripts?